关键词: 认知增强, 机器学习, 脑机融合

Abstract: Cognitive ability refers to the capacity to process, store, and retrieve information, and it is a strong predictor of short- and long-term achievements such as academic performance, social skills, health, wealth, and involvement in criminal activities. Common cognitive enhancement methods include cognitive training, neurofeedback, and electrical stimulation. However, these methods have several shortcomings, including long training sessions, fixed training parameters, and inconsistent training outcomes. Recently, machine learning has provided an opportunity to obtain more intelligent, personalized, and precise cognitive training solutions, which can overcome the shortcomings of previous methods and maximize training effectiveness. Therefore, this study first introduced common cognitive enhancement and machine learning methods. Then, the relationship between machine learning and cognitive enhancement was analyzed, and a comprehensive overview of the application of machine learning in cognitive enhancement was provided. Finally, the challenges and potential directions for future research were discussed.
Cognitive training is the most common cognitive enhancement method that can effectively improve the performance on training tasks and structurally similar untrained tasks. Meanwhile, successful cognitive training may potentially enhance the performance on transfer tasks and show maintenance effects (i.e., cognitive control, reasoning, intelligence, and cross-modal tasks). However, cognitive training can be time-consuming and has inconsistent effectiveness. Neurofeedback enables individuals to self-regulate their brain activity through signals such as electroencephalography (EEG), magnetic resonance imaging (MRI), and near-infrared spectroscopy (NIRS), and can significantly improve attention, working memory, and cognitive control. However, neurofeedback equipment is expensive, training parameters may be overly simplified, and the training process is affected by emotions. Electrical stimulation techniques are used to activate or inhibit specific brain regions, such as the dorsolateral prefrontal cortex. Studies have shown that both single and repetitive stimulation can improve executive control, attention, and multitasking abilities. Moreover, the combination of electrical stimulation with cognitive training can further improve training outcomes. However, the underlying mechanism is still unclear, and problems such as fixed parameters and inconsistent training effects remain.
Supervised machine learning is extensively used in the fields of neuroscience and psychology. Specifically, algorithms and labeled datasets are used to train machine learning models that can recognize patterns, and then these models are used to predict labels for new data. Supervised machine learning can be divided into regression algorithms, classification algorithms, and deep learning. Specifically, regression algorithms predict the scores of psychological traits by outputting continuous values, including linear regression, stepwise regression, lasso regression, ridge regression, elastic net regression, and support vector regression. For classification tasks such as determining whether an individual has a psychological disorder, algorithms like logistic regression, linear discriminant analysis, k-nearest neighbors, support vector machines, decision trees, and random forests are adopted. Additionally, deep learning models such as convolutional neural networks, recurrent neural networks, and transformer networks are particularly suitable for complex cognitive tasks because of their ability to automatically learn feature representations.
There is a mutually beneficial relationship between machine learning and cognitive enhancement. On the one hand, machine learning helps to overcome the limitations of cognitive enhancement methods and maximize the effectiveness of training. Specifically, machine learning can assist in the selection of difficulty levels and training programs before the training. Meanwhile, machine learning can be employed to adjust the task difficulty and electrical stimulation parameters during the training process. Another scenario involving machine learning during the training process is neurofeedback decoding, where machine learning is utilized to decode task-related brain activity patterns, which are then used as indicators for implementing precise real-time neurofeedback. As for training effects, machine learning can be adopted to analyze the relationship between brain data and cognitive abilities, which helps to reduce the number of assessment tasks. On the other hand, the integration and enhancement of brain signals can improve the performance of machine-learning models. This brings a new meaning to cognitive training from the perspective of machine learning.
It is noteworthy that related research is still in the exploratory stage and faces several challenges. First, in terms of training evaluation, existing cognitive prediction models have relatively low accuracy, and this issue can be addressed in future research through algorithm selection, multimodal fusion, feature selection, and hyperparameter tuning. Second, sample diversity is crucial for personalized cognitive training and assessment. Future research should focus on increasing the sample size and diversity of publicly available datasets. Third, model interpretability can help researchers to understand the mechanisms underlying cognitive training. Therefore, future research should flexibly apply interpretability methods in traditional machine learning models and develop interpretable deep learning methods to ensure the interpretability of deep learning models.

Key words: cognitive enhancement, machine learning, brain-computer fusion